• Aucun résultat trouvé

From bi-ideals to periodicity

N/A
N/A
Protected

Academic year: 2022

Partager "From bi-ideals to periodicity"

Copied!
9
0
0

Texte intégral

(1)

DOI:10.1051/ita:2008010 www.rairo-ita.org

FROM BI-IDEALS TO PERIODICITY

J¯ anis Buls

1

and Aivars Lorencs

2

Abstract. The necessary and sufficient conditions are extracted for periodicity of bi-ideals. They cover infinitely and finitely generated bi-ideals.

Mathematics Subject Classification.68R15, 94A55, 68Q15.

1. Introduction

The periodicities are fundamental objects, due to their primary importance in word combinatorics [8,9] as well as in various applications. The study of periodicities is motivated by the needs of molecular biology [6] and computer science. Particularly, we mention here such fields as string matching algorithms [4], text compression [13] and cryptography [11].

In different areas of mathematics, people consider a lot of hierarchies which are typically used to classify some objects according to their complexity. Here we deal with the hierarchy

BP, where Bis the class of bi-ideals,

Pis the class of periodic words.

This hierarchy comes from combinatorics on words, where these classes are being investigated intensively (cf. [2,8–10]). Bi-ideal sequences have been considered, with different names, by several authors in algebra and combinatorics [1,3,7,12,14].

Every bi-idealx is the limit of some bi-ideal sequence (vi). This bi-ideal se- quence can be represented uniquely by the sequence (ui), where v0 = u0 and

Keywords and phrases.Periodic words, bi-ideals, the sequence generates the bi-ideal, finitely generated bi-ideals.

1 Department of Mathematics, University of Latvia, Rai¸na bulv¯aris 19, R¯ıga, 1586, Latvia;

[email protected]; web site: http://home.lanet.lv/~buls

2 Institute of Electronics and Computer Science, Dz¯erbenes street 14, R¯ıga, 1006, Latvia;

[email protected]

Article published by EDP Sciences c EDP Sciences 2008

(2)

∀i≥0vi+1=viui+1vi. We characterize the periodic words through this represen- tation. At first we give an exhaustive description (Th. 3.7) of periodicity for all classes of bi-ideals. Then for periodic bi-ideals we demonstrate if everyui appears infinitely often then every ui is a power of the certain word. This leads to the effective method for finitely generated bi-ideals to check whether the bi-ideals are periodic.

2. Preliminaries

In this section we present most of the notations and terminology used in this paper. Our terminology is more or less standard (cf. [10]) so that a specialist reader may wish to consult this section only if need arise.

LetAbe a finite non-empty set andA the free monoid generated byA. The set Ais also called analphabet, its elementsletters and those of A finite words.

The role of the identity element is performed by theempty wordwhich is denoted byλ. We setA+=A\{λ}.

A word w A+ can be written uniquely as a sequence of letters as w = w1w2. . . wl, with wi∈A, 1≤i≤l, l >0. The integerl is called the lengthofw and denoted|w|. The length of λis 0. We setw0=λand∀i wi+1=wiw;

w+= i=1

{wi}, w =w+∪ {λ}.

A positive integerpis called aperiodofw=w1w2. . . wlif the following condition is satisfied:

1≤i≤l−p wi=wi+p.

We recall the important periodicity theorem due to Fine and Wilf [5]:

Theorem 2.1. Let w be a word having periods pand q and denote by gcd(p, q) the greatest common divisor ofpandq. If|w| ≥p+q−gcd(p, q), thenwhas also the periodgcd(p, q).

The word w A is called a factor (or subword) of w A if there exist u, v A such that w = uwv. The word u (respectively v) is called a prefix (respectively asuffix) ofw. The ordered triple (u, w, v) is called anoccurrenceof w inw. The factorw is called aproperfactor ifw=w. We denote respectively byF(w), Pref(w) and Suff(w) the sets ofwfactors, prefixes and suffixes.

An (indexed) infinite wordxon the alphabet A is any total mapx : N→A. We set for anyi≥0,xi=x(i) and write

x= (xi) =x0x1. . . xn. . . The set of all the infinite words overAis denoted byAω.

The word w A is a factor of x∈ Aω if there exist u∈ A, y Aω such thatx=uwy. The wordu(respectivelyy) is called aprefix(respectively asuffix)

(3)

of x. We denote respectively byF(x), Pref(x) and Suff(x) the sets ofxfactors, prefixes and suffixes. For any 0≤m≤n, bothx[m, n] andx[m, n+ 1) denote a factorxmxm+1. . . xn. The indexed wordx[m, n] is called anoccurrenceofw inx ifw =x[m, n]. The suffixxnxn+1. . . xn+i. . .is denoted byx[n,∞).

Ifv ∈A+ we denote byvω the infinite wordvω =vv . . . v . . .This wordvω is called aperiodic word. The concatenation ofu=u1u2. . . uk ∈A andx∈Aω is the infinite word

ux=u1u2. . . ukx0x1. . . xn. . .

A word xis called ultimately periodic if there exist words u∈A, v ∈A+ such thatx=uvω. In this case,|u|and|v|are called, respectively, an anti-periodand aperiodofx.

A sequence of words ofA

v0, v1, . . . , vn, . . .

is called a bi-ideal sequence if ∀i 0 (vi+1 viAvi). The term “a bi-ideal sequence” is due to the fact that ∀i≥0 (viAvi) is a bi-ideal ofA.

Corollary 2.2. Let (vn)be a bi-ideal sequence. Then vmPref(vn)Suff(vn) for allm≤n.

A bi-ideal sequencev0, v1, . . . , vn, . . .is calledproperifv0. In the following the term bi-ideal sequence will be referred only to proper bi-ideal sequences.

Ifv0, v1, . . . , vn, . . .is a bi-ideal sequence, then there exists a unique sequence of words

u0, u1, . . . , un, . . . such that

v0=u0, ∀i≥0 (vi+1=viui+1vi).

Let us consideru, v∈A=A∪Aω. Thend(u, v) = 0 ifu=v, otherwise d(u, v) = 2−n,

where

n= max{ |w| |w∈Pref(u)Pref(v)}. It is called aprefix metric.

Let v0, v1, . . . , vn. . . be an infinite bi-ideal sequence, where v0 = u0 and

∀i 0 (vi+1 = viui+1vi). Since for all i 0 the word vi is a prefix of the next wordvi+1 the sequence (vi) converges, with respect to the prefix metric, to the infinite wordx∈Aω

x=v0(u1v0)(u2v1). . .(unvn−1). . .

This wordxis called abi-ideal. We say the sequence (ui)generatesthe bi-idealx.

(4)

Convention. Let x be a bi-ideal generated by (ui), then x = lim

i→∞vi, where v0=u0 andvi+1=viui+1vi. We adopt this notational convention henceforth.

Letxbe an infinite word. A factoruofxis calledrecurrentif it occurs infinitely often inx. The wordxis calledrecurrentwhen any of its factors is recurrent.

Proposition 2.3. (see, e.g., [10]) A word x is recurrent if and only if it is a bi-ideal.

Lemma 2.4. (see, e.g., [10]) Let x∈Aω be an ultimately periodic word. Ifxis recurrent, then xis periodic.

Due to this lemma we can restrict ourselves. Therefore we investigate only the periodicity of bi-ideals and say nothing about ultimate periodicity.

3. The periodicity of bi-ideals

The following three lemmas are very easy, but they turn out to be extremely useful:

Lemma 3.1. If x=wω and T is the minimal period of the word x, then T\|w|, i.e. T divides |w|.

Proof. Letn=T|w|, then bothT and|w|are periods of the wordx[0, n). Hence (Th.2.1)t= gcd(T,|w|) is a period ofx[0, n). Now we have

∀i x[0, n) =x[ni, n(i+ 1)).

Thereforet is a period ofx. Since T is the minimal period of the word x, then t≥T gcd(T,|w|) =t. HenceT = gcd(T,|w|), therebyT\|w|.

Lemma 3.2. If x=wω=uvyand|w|=|v|, thenvy=y=vω.

Proof. Let|w| =t and |u| =k+ 1, then v =xk+1xk+2. . . xk+t, since |v| =|w|.

We have∀i xi+t=xi, therefore

∀j∈1, t∀s xk+j=xk+j+st.

Lemma 3.3. If ∃u∈ A+ ux = x∈ Aω, then a word x is periodic with the minimal periodT\|u|.

Proof. Letu=a1a2. . . at−1, where∀j aj∈A, and y=ux, then

∀i xi=yi+t. Let

y=ux=x . Hence

∀i yi =xi=yi+t.

This means thatyis periodic with a periodt. Sincey=x, thenxis periodic with a period t too. Let T is the minimal period of x, then by Lemma 3.1 T\t, i.e.

T\|u|.

(5)

Corollary 3.4. Let |v| be the minimal period ofx=vω. If v=x[k, k+|v|) then |v|\k.

Proof. If, for any k,v=x[k, k+|v|), then (see Lem.3.2) x=x[0, k)vω=x[0, k)x.

Hence by Lemma3.3|v|\|x[0, k)|=k.

Lemma 3.5. If existsn such thatvnu∈v and∀i∈Z+(un+i∈uv), then

∀i∈N(vn+i∈vvn). Proof. Ifi= 0 thenvn+i=vn=λvn∈vvn.

Further, we shall prove the lemma by induction on i, i.e., suppose that vn+i ∈vvn, namely,

∃k∈N(vn+i=vkvn). By assumption,vnu∈v andun+i+1∈uv,i.e.

∃l∈N(vnu=vl) ∧ ∃m∈N(un+i+1=uvm). Hence

vn+i+1 = vn+iun+i+1vn+i= (vkvn)(uvm)(vkvn)

= vk(vnu)vm+kvn=vkvlvm+kvn∈vvn.

We have completed the inductive step.

Lemma 3.6. If tis the period of the bi-ideal xand|vn| ≥t, then

∀i∈Z+un+1x=un+ix .

Proof. We havevn+i=vn+i−1un+ivn+i−1. Hence, ifi∈Z+ then (Cor.2.2)

∀i∈Z+∃vi vn+i =vnvivn. Now, by definition ofx

x = vnun+1vn. . .

x = vn+iun+i+1vn+i. . .=vnvivnun+i+1vn. . . By assumption,xis periodic, therefore

x=vω, where |v|=t .

(6)

Sincev∈Pref(vn) then by Lemma3.2

x = vnun+1x , x = vnun+i+1x .

Hence∀i∈Z+x=vnun+ix .Thus∀i∈Z+ un+1x=un+ix . Theorem 3.7. A bi-idealxis periodic if and only if

∃n∈N∃u∃v(vnu∈v ∧ ∀i∈Z+un+i∈uv).

Proof. LetT be the minimal period of the wordx, then∃n∈N|vn| ≥T. Thus by Lemma3.6

∀i∈Z+ un+1x=un+ix . Letube the longest word of the set

i=1Pref(un+i) then

∀i∈Z+∃ui(un+i=uui). Particularly,∃k un+k=u. This means that

∀i∈Z+ uuix=un+ix=un+kx=ux . Thus

∀i∈Z+ uix=x . Hence by Lemma3.3

∀i∈Z+ T\|ui|. Thereby

∀i∈Z+ ui∈v, wherev=x[0, T). Thus

∀i∈Z+ un+i=uui∈uv. Note

x=vnun+1vn. . .=vnuu1vn. . .

Sinceu1 ∈v and v Pref(vn), then [Lemma3.2] x=vnux. Hence [Lem.3.3]

vnu∈v.

By Lemma3.5

∀i∈N∃kiNvn+i=vkivn. Since lim

k→∞|vk|=then lim

i→∞ki=∞. Thus x= lim

k→∞vk = lim

i→∞vn+i= lim

i→∞vkivn =vω.

(7)

4. Powers

Observation. If allui∈w for some wordw =λ, then the bi-ideal generated by (ui) is periodic.

The following example demonstrates the converse is not true in general.

Example 4.1. Letxbe the bi-ideal generated by (ui), where u0 = 0,

u1 = 1,

∀i >1 ui = 00100. Then

v0 = 0, v1 = 010,

v2 = 010 00100 010,

v3 = 01000100010 00100 01000100010, . . .

andx= lim

i→∞vi= (0100)ω. Thusxis periodic.

Nevertheless, if every uj appears infinitely often in (ui), then the converse is valid.

Theorem 4.2. Let(ui)be a sequence of words, which contains everyuj infinitely often. The bi-idealxgenerated by (ui)is periodic if and only if

∃w∀i ui∈w.

Proof. Letxbe a periodic bi-ideal, then by Theorem3.7

∃n∈N∃u∃v(vnu∈v ∧ ∀i∈Z+un+i∈uv).

Hence by Lemma3.5 |v|is the period ofx. Therefore we can assume that |v|is the minimal period ofxand|u|<|v|. Since the sequence (ui) contains everyuj

infinitely often then by Theorem3.7 ∀i∈N(ui ∈uv).

Now suppose thatui=ufor alli < mbutum=uvk, wherek >0. Then there existα∈Z+ andy such that

x=uαvky . (i) Ifu=λthen∀i ui∈v.

(ii) Otherwiseu=λ. Then (Corollary3.4)|v|\α|u|. Hence, there existsβ∈Z+

such thatα|u|=β|v|. Thusx=vω=uω. Contradiction, since|u|<|v|and|v|is the minimal period ofx.

See Observation.

(8)

Now we turn our attention to the problem of effectiveness.

Definition 4.3. Assume that (ui) generates a bi-idealx. The bi-idealxis called finitely generatedif

∃m∀i∀j(i≡j(modm)⇒ui =uj).

In this situation, we say that the m–tuple (u0, u1, . . . , um−1) generates the bi- idealx.

Theorem 4.4. A bi-ideal x generated by (u0, u1, . . . , um−1) is periodic if and only if

∃w∀i∈0, m−1 ui∈w.

Proof. As a corollary from Definition4.3and Theorem4.2.

This theorem gives a method to generate nonperiodic bi-ideals. Let (u0, u1, . . . , um−1)

be anym-tuple chosen at random. Letv be any shortest word from the set {u0, u1, . . . , um−1}

andw be the shortest prefix of v such thatv ∈w+. If there exists ui such that ui∈/ wthen the bi-ideal generated by (u0, u1, . . . , um−1) is not periodic. This can be easily checked by a deterministic algorithm.

Acknowledgements. The useful suggestions of two referees are gratefully acknowledged.

References

[1] D.B. Bean, A.E. Ehrenfeucht and G. McNulty. Avoidable patterns in strings of sym- bols.Pacific J. Math.85(1979) 261–294.

[2] J. Berstel, J. Karhum¨aki.Combinatorics on Words — A Tutorial.TUCS Technical Report (No. 530, June) (2003).

[3] M. Coudrain and M.P. Sch¨utzenberger. Une condition de finitude des mono¨ıdes fin- iment engendr´es.C.R. Acad. Sci. Paris, S´er. A262(1966) 1149–1151.

[4] M. Crochemore and W. Rytter. Squares, cubes, and time-space efficient string searchinng.Algorithmica13(1995) 405–425.

[5] N.J. Fine, H.S. Wilf. (1965) Uniqueness theorem for periodic functions.Proc. Amer.

Math. Soc.16(1965) 109–114.

[6] D. Gusfield. Algorithms on Strings, Trees, and Sequences. Cambridge University Press (1997).

[7] N. Jacobson. Structure of Rings. American Mathematical Society, Providence, RI (1964).

[8] M. Lothaire.Combinatorics on Words.Encyclopedia of Mathematics and its Appli- cations, Vol. 17. Addison-Wesley, Reading, Massachusetts (1983).

(9)

[9] M. Lothaire.Algebraic Combinatorics on Words.Encyclopedia of Mathematics and its Applications, Vol. 90. Cambridge University Press, Cambridge (2002).

[10] Aldo de Luca, Stefano Varricchio.Finiteness and Regularity in Semigroups and For- mal Languages.Springer-Verlag, Berlin, Heidelberg (1999).

[11] R.A. Rueppel. Analysis and Design of Stream Ciphers. Springer-Verlag, Berlin (1986).

[12] I. Simon. Infinite words and a theorem of Hindman. Rev. Math. Appl. 9 (1988) 97–104.

[13] J.A. Storer. Data compression: methods and theory. Computer Science Press, Rockville, MD (1988).

[14] A.I. Zimin. Blocking sets of terms.Matem. sb.,119, 363–375(Russian) (1982).

Références

Documents relatifs

ompute the minimal dierential prime omponents of regular ideals and provides also.. informations about these prime ideals without having to ompute them ; the

The second argument of the proof is an explicit upper bound on the length of ascending chains of polynomial ideals generated by adding consecutive derivatives, given in Theorem 4

Also, W — 82 is a closed semi analytic set in C 2 of dimension at most one, and can hence be written as So u Si where So is a finite set of points and Si is a relatively

This paper defines and studies the notion of primitive root of a bi-periodic infinite picture, that is a rectangular pattern that tiles the bi-periodic picture and contains exactly

The authors proved a very interesting result ( [7, Theorem 1.17]): a reduced ring R is Noetherian if each of its prime ideals has a finitely generated power.. This question 0.1

It is also proven that each commutative ring R whose finitely generated ideals are quasi-flat is of λ -dimension ≤ 3, and this dimension ≤ 2 if R is local.. This extends a former

— There is no exchange ring (thus, no von Neumann regular ring and no C*-algebra of real rank zero) with finite stable rank whose semilattice of finitely generated,

A list decodable code is an error correcting coding scheme such that, if the codeword z for a string x is known without too many errors, one can produce a small list of strings to